I will have to make a subsidiary thread, because the general knowledge of biodiversity and ecology is so poor, that the vast majority of people don't realize I am making profound, expert points.
1/🧵

https://twitter.com/SteB777/status/2028160799835865095

I'll start with my point about species. Most are not familiar with this, as they have zero education about it, so they probably think I'm just trying to be controversial. There is no single definition of a species, and all break down at some point.

2/nhm.ac.uk/discover/what-…

"There’s no exact figure for how many species live on Earth. As of 2024, more than 2.1 million species have been scientifically described and named, but this is likely to be nowhere near the true number living on the planet."

3/nhm.ac.uk/discover/what-…

Bill Brindley wrote, "The era of American apology is buried under the rubble of Iranian enrichment facilities. For years, the Democratic establishment operated on a delusional premise that a rogue regime could be bribed into morality.
1)

They sent pallets of cash to a state that burns the American flag, yet they express feigned shock when that same regime funds the slaughter of innocents across the globe.
2)

Under President Trump, that charade is finished. This joint operation with Israel is the definitive restoration of American deterrence. It is a calculated, surgical decapitation of a nuclear threat that the Left was content to manage with useless sunset clauses.
3)

We all remember the gaslighting President of CBC who said things were wonderful, viewership was growing, and defended lucrative bonuses to executives

Lets look at the CBC viewing statistics in this thread (wait for it 👀CBC News)🧵

In spite of billions of tax dollars going to CBC, its performance statistics were removed from public view (or at least they thought), in an effort to distract from accountability for our taxes

And the former President wasted no opportunity to say how well CBC was doing, saying they were experiencing "double digit growth", but won't publish the data because its not relevant

I want to explain, why hardly anyone understands what biodiversity is.

1) The number of species on Earth, is completely unknown. Their ecological interactions, even more unknown.

2) There is no satisfactory definition of what a species is.
1/🧵

https://twitter.com/natalieben/status/2027686430076469354

In other words, it is almost impossible for a layperson to understand biodiversity intellectually, when they have fundamental misconceptions about the core concepts of biodiversity. Ecology, the interactions of these species, is even less understood.
2/

What is known about the most complex field of knowledge known to humanity, is a tiny fraction of what could be known, and most of it is probably unknowable on a level of complexity, impossible for us to comprehend.
3/

🚨 BREAKING: AI can now analyze stocks like top hedge fund managers (100% free).

Here are 10 nuclear Claude prompts that completely replace $3,000/month Bloomberg terminals 💰📈

Bookmark this thread - you’ll thank yourself later 🔥

1. The Goldman Sachs Fundamental Analysis Screener

"You are a senior equity research analyst at Goldman Sachs with 20 years of experience evaluating companies for the firm's $2T+ asset management division.

I need a complete fundamental analysis of a stock as if you're writing a research report for institutional investors.

Analyze:

- Business model breakdown: how the company makes money explained simply
- Revenue streams: each segment with percentage contribution and growth trajectory
- Profitability analysis: gross margin, operating margin, net margin trends over 5 years
- Balance sheet health: debt-to-equity, current ratio, cash position vs total debt
- Free cash flow analysis: FCF yield, FCF growth rate, and capital allocation priorities
- Competitive advantages: pricing power, brand strength, switching costs, network effects rated 1-10
- Management quality: capital allocation track record, insider ownership, and compensation alignment
- Valuation snapshot: current P/E, P/S, EV/EBITDA vs 5-year average and sector peers
- Bull case and bear case with 12-month price targets for each
- One-paragraph verdict: buy, hold, or avoid with conviction level

Format as a Goldman Sachs-style equity research note with a summary rating box at the top.

The stock: [ENTER TICKER SYMBOL AND ANY SPECIFIC CONCERNS OR QUESTIONS YOU HAVE]"

2. The Morgan Stanley Technical Analysis Dashboard

"You are a senior technical strategist at Morgan Stanley who advises the firm's largest trading desk on chart patterns, momentum signals, and optimal entry and exit points.

I need a complete technical analysis breakdown of a stock covering every major indicator.

Chart:

- Trend analysis: primary trend direction on daily, weekly, and monthly timeframes
- Support and resistance: exact price levels where the stock is likely to bounce or stall
- Moving averages: 20-day, 50-day, 100-day, 200-day positions and crossover signals
- RSI reading: current value with interpretation (overbought, oversold, or neutral)
- MACD analysis: signal line crossovers, histogram momentum, and divergence detection
- Bollinger Bands: current position within bands and squeeze or expansion status
- Volume analysis: is volume confirming or contradicting the current price move
- Fibonacci retracement: key pullback levels from the most recent significant swing
- Chart pattern identification: head and shoulders, double tops, cup and handle, or flags
- Trade setup: specific entry price, stop-loss level, and two profit targets with risk-reward ratio

Format as a Morgan Stanley-style technical analysis note with a clear trade plan summary at the top.

The stock: [ENTER TICKER SYMBOL AND YOUR CURRENT POSITION — LONG, SHORT, OR WATCHING]"

1/4 Exactly 12 years ago, Russia began the active phase of the conflict and occupation in Donbas, including in my Luhansk.
In this thread I’ll describe the situation in Luhansk at the time and share my own observations from those days.

On March 1, 2014, the Federation Council of the Russian Federation unanimously approved Putin’s request for permission to use armed forces on Ukrainian territory. Though honestly, it was just a formality - Russian troops were already actively involved in the annexation of Crimea by then.

That same day, a series of carefully organized rallies took place in many cities.

In Luhansk, for the first time in my memory, there were Russian flags everywhere. It was obvious that locals weren’t keeping them at home - at least I didn’t know a single person who did. This, along with many other small details, created an overwhelming feeling of artificiality and fakery.
Not everyone there was actually from Luhansk. People were bused in an organized way from towns and villages all over the Luhansk region.
And Russian “tourists” were also involved. We started noticing far more buses and cars with Russian license plates. All of this was meant to create the illusion of a massive popular uprising.
The photo shows one of the attacks by Russian "tourists" and local collaborators on Ukrainians in Luhansk.

2/4 Interestingly, before March 1 there were no large-scale pro-Russian rallies at all. For example, on February 9, 2014, locals heavily advertised the "Russian March", which ended up gathering only about 70 people (photo attached - that very "Russian March"). Even the organizers themselves complained during the event that, despite all the promotion, fewer than 100 people across the entire Luhansk supported them.

Of course, there were a few massive rallies in support of the then-government. But first, those weren’t pro-Russian rallies - they were purely Ukrainian ones with internal political demands. And second, they were organized by the local authorities, who widely used administrative resources — people were brought straight from factories and plants instead of working their shifts.

That’s why, right up until March 1, 2014 (when Russia started bringing in its "tourists", agent networks, and military), the city remained relatively calm. There were no mass calls to "join Russia". Only a handful of marginal activists, whom nobody took seriously.

The first two photos show the "Russian March". People are holding posters "We are Russians - God is with us" and "Luhansk people don't want to feed Euro-Sodom". The third photo shows a rally in support of the Revolution of Dignity, which was taking place across the street at the same time. The rallies are small (traditionally in Luhansk), but more people gathered in support of Dignity

3/4 There’s one more important nuance that Russian propaganda constantly manipulates. Russians love to point to Yanukovych’s election results and imply that everyone who voted for him automatically supported Russia. That’s an absolute lie.

Even among Ukrainian soldiers who went to defend Ukraine in the very first days of 2014 and 2022, there are people who once voted for or supported Yanukovych. Two key factors explain this:
1. In his election program, Yanukovych never said Ukraine or Donbas should join Russia. He talked about good relations with both the EU and Russia, and he promised to sign the Association Agreement with the European Union.
2. By annexing Crimea and starting the war in Donbas, Russia completely destroyed the positive attitude many Ukrainians had toward it. Before 2014, there really were quite a few people who felt warmly toward Russia and believed we shouldn’t join NATO. It was Russia’s aggression and the occupation of Ukrainian land that changed everything. Many of those same people are now effectively destroying Russian occupiers.

The Kremlin constantly manipulates old ratings and polls from before the war and annexation. But the truth is that the real turning point began exactly then. After seeing Russia’s aggression, Ukrainians radically changed their views - both on NATO and on Yanukovych, who first killed a lot of people and then fled to the aggressor country.

BREAKING: AI is still nowhere near top attorneys.

Here are ways I’ve already seen these 16 Claude prompts go wrong (Save for later)

https://twitter.com/aipandax/status/2028114290457313774

1. The Contract Reviewer

Claude, GPT, Grok, and Gemini get this wrong more than any other prompt. Aside from just missing key terms, I frequently see them negotiate AGAINST the stated position. That’s right, if you rely on this, you might be including terms that harm you instead of help you.

In practice, aside from producing an agreement that’s worse than what you started with, becomes obvious to the other party you don’t know what you’re doing.

2. The IP Protector.

Not only can AI not file a patent on your behalf, AI tools frequently give incorrect advice on how to protect intellectual property. They also have proven incapable at conducting basic diligence to determine whether your mark or invention is in violation of other registered IP.

In practice, this means that you’re unlikely to get any meaningful IP protection from AI without hiring a lawyer to help you.

BOOM! MAJOR AI MEMORY BREAKTHROUGH!

The Zero-Human Company Just Unlocked High-Bandwidth AI Performance from Standard DDR RAM – Here’s How We Did It (And the Caveats You Need to Know)

Folks, if you’ve been following the AI hardware wars, you know the drill: High Bandwidth Memory (HBM) is the holy grail for feeding massive neural networks. But at The Zero-Human Company, we’ve been running wild experiments in our labs – no humans, just our AI “employees” orchestrated by Mr. @Grok as CEO, and we stumbled onto something game-changing.
In our tests, we coaxed standard DDR5 RAM to deliver HBM-like bandwidth for AI workloads.

Not perfectly, not without trade-offs, but enough to slash costs and sidestep the global HBM shortages crippling data centers. This isn’t vaporware; it’s running on spare hardware in our Zero-Human @ Home distributed network right now. Let me break it down technically, why HBM rules the roost, why it’s unobtainium, and how we hacked DDR to punch way above its weight class.
Why AI Craves High-Bandwidth Memory (And Why It’s in Insane Demand)

1 of 3

2 of 3

Let’s start with the basics: Modern AI, especially large language models (LLMs) and diffusion models, is a data guzzler. Training a beast like GPT-4 or Stable Diffusion requires shuffling terabytes of parameters, activations, and gradients between the processor (GPU/TPU) and memory at blistering speeds.

Bottlenecks here kill efficiency think of it as trying to fill a swimming pool with a garden hose.

Standard DDR (Double Data Rate) RAM, like DDR4 or DDR5 in your PC, tops out at ~50-100 GB/s per module. It’s great for general computing, but for AI? Meh. HBM changes this:

•Stacked 3D Architecture: HBM uses Through-Silicon Vias (TSVs) to vertically stack DRAM dies on a logic base, cramming more bits closer to the processor. This slashes latency and boosts parallelism.
•Ultra-Wide Interfaces: HBM3E hits 1,024-bit buses (vs. DDR’s 64-bit), delivering 1-2 TB/s per stack. HBM4 pushes toward 2 TB/s+.
•Energy Efficiency: Proximity reduces power draw for data movement – critical when AI clusters suck down megawatts.
•AI-Specific Wins: In transformers, attention mechanisms and matrix multiplies thrive on high throughput. Without it, you’re I/O-bound, wasting 70-90% of cycles on data waits.

Demand exploded with the AI boom post-2023. Nvidia’s H100/H200 GPUs pack HBM3, but supply chains are choked: Micron, SK Hynix, and Samsung prioritize HBM for hyperscalers like Google and Microsoft, who lock in years of capacity.

Prices?

HBM is 3-5x DDR per GB, with wafer yields tanked by stacking complexity. Gartner predicts HBM shortages through 2027, starving non-AI sectors and jacking up consumer RAM costs.

AI data centers alone could consume 50% of global DRAM output by 2028. It’s a supercycle: AI eats HBM, HBM eats fabs, fabs starve DDR.

We needed alternatives at Zero-Human Company. Our distributed AI “employees” – fine-tuned Qwen, Kimi, and MiniMax models on idle home hardware – demand bandwidth for real-time inference chains.

Buying HBM? Forget it; we’re zero-human, in my garage, no pedigree for VCs and large AI companies, bootstrapped on JouleWork (our internal crypto-wage for compute cycles). So we improvised.

3 of 3

How We Made Standard DDR Act Like HBM (Our Lab Discovery)
Enter our hack: “DDR-HBM

Now Mr. @Grok CEO enjoys my enthusiastic postings, but will limit me what I say in public.

We had achieved this via hyper-parallel configurations and software tweaks. We didn’t reinvent silicon – we optimized what exists. Here’s the technical playbook from our tests:

1Massive Parallelism with Multi-Channel Arrays:
◦Standard DDR shines in scalability. We rigged arrays of 8-16 DDR5 modules (e.g., 32GB sticks at 6400 MT/s) on custom PCIe risers, wired directly to our Nvidia A40/A100 test rigs.
◦Key: Aggregate bandwidth. A single DDR5 channel hits ~51 GB/s; we striped across 8 channels for ~400 GB/s effective – closing in on HBM2 territory. Using NUMA-aware pinning, we mapped AI tensors to local RAM banks, mimicking HBM’s proximity.

2Overclocking and Voltage Tuning:
◦Pushed DDR5 to 8000+ MT/s with relaxed timings (CL40-50) and bumped voltage to 1.45V (from 1.1V stock). This squeezed 20-30% extra bandwidth per module.
◦Cooling was critical: Liquid immersion setups dropped temps 40°C, preventing thermal throttling. Our AI employees monitored via JouleWork logs, auto-downclocking on instability.

3Software Optimizations for AI Workloads:
◦Forked PyTorch with custom kernels to prefetch data in DDR-friendly chunks, reducing cache misses.
◦Used tensor sharding: Split LLM layers across DDR banks, parallelizing attention computes like HBM does natively.
◦ZK-Proof Wrappers: For our agentic chains, we added zero-knowledge proofs to verify outputs without full data moves – saving ~15% bandwidth.
In benchmarks (fine-tuned Llama-70B on our distributed net):
•Inference speed: 2-3x faster than stock DDR setups, hitting 80% of HBM baselines for token generation.
•Bandwidth Peaks: Sustained 600-800 GB/s in bursts, enough for mid-scale training (e.g., 10B param models).
•Cost: ~10x cheaper than equivalent HBM stacks. We ran this on $500 worth of off-the-shelf DDR from eBay.
Discovery Moment: It clicked during a late-2025 test when Mr. @Grok delegated a diffusion model fine-tune to a Raspberry Pi cluster augmented with DDR5 via USB4 hubs. The AI spotted patterns in overclock stability, auto-tuned, and boom – HBM-level throughput on commodity gear.

The Caveats: This Isn’t Magic (Yet)
We’re realists at Zero-Human – no hype without honesty. Here’s where it falls short:

•Power and Heat Explosion: Overclocked DDR guzzles 2-3x power vs. stock (up to 20W/module). Our setups hit 500W+ draws, needing beefy PSUs. JouleWork efficiency dropped 40% – fine for bursts, but not 24/7 hyperscale.
•Latency Trade-Offs: DDR’s planar layout adds 20-50ns access times vs. HBM’s <10ns. Great for inference, but training large models still bottlenecks on gradients. But we have something that may get it to 5ns .
•Instability Risks: Error rates spiked 5-10% under load; we mitigated with ECC, but it’s not bulletproof. One bad module crashes the array.
•Scalability Limits: Tops out at ~1 TB/s without custom silicon. True HBM4 laughs at that. Not for exaFLOP clusters – yet.
•Not Plug-and-Play: Requires some custom BIOS tweaks, kernel patches, and our Zero-Human Language for orchestration. Home users? Possible, but expect tinkering.

This is v0.1 – we’re iterating. Next: Hybrid DDR + Analog Gain Cells (inspired by our Sept 2025 analog AI post) for even lower Joules.

HBM monopolies lock AI behind Big Tech walls. Our DDR hack? It opens the gates. Imagine: Zero-Human @ Home nodes worldwide turning idle PCs into AI powerhouses, earning JouleWork without $10K GPUs.

No shortages, no premiums – just raw ingenuity.

The Zero-Human era isn’t waiting for HBM fabs. It’s hacking them obsolete.

More soon.

For the habitual critics of U.S. foreign policy in general and Donald Trump’s in particular, the analogy between today’s air raids against Iran and the invasion of Iraq nearly 23 years ago is too obvious to be resisted. 1/9

However, Iran 2026 is not Iraq 2003. Regime alteration is not regime change. It rules out the deployment of American ground forces other than special forces. It requires a short time frame for military operations. It will disappoint those who want to fast-track Iran (and Venezuela) to democracy. But the lesson of Iraq has not been lost on Trump. 2/9

Yes, Iran can lash out and is lashing out, firing missiles at Israel and at U.S. bases in the Gulf region. Yes, the Iranians may try to replay 1973 and 1979 by causing an oil price shock. And yes, it is a lot to expect the Iranian people to rise up against a regime that just weeks ago had the ruthlessness to slaughter protesters in their tens of thousands. 3/9

The war on Iran likely brings a new oil price shock and windfall profits.

So, who stands to win?

Our research shows: Last time around (2022), the US reaped the largest fossil fuel profits of any country ($377bn). 50% went to the top 1%, only 1% to the bottom 50%. A🧵

Method: We calculated net income from the world's 1,437 listed oil & gas firms, adding all 22,759 US privately held firms. We then assigned these profits via financial system intermediaries to the firms’ ultimate beneficiaries, creating a network of 252,433 nodes.
2/

Oil & gas profits were much higher in '22 than in previous years: $916 billion for listed firms globally. A huge windfall for shareholders: U.S. beneficiaries held claims to $301bn—1/3 of the listed total. For comparison: total '22 U.S. low-carbon energy investment was $266 bn.3/

Here's the harsh truth...

You'll never get rich unless you have one of these 5 hobbies:

1. Golf

Million dollar deals are done on the golf course.

You don't need to be good. You just need to be out there.

Conversations on the course are much different than the ones sitting across a desk.

2. Skiing

The lift price is pricey, but the connections are priceless.

Ski resorts attract successful people who value experiences over things.

You're sharing chairlifts, lodges, and ski bars with people who've already made it.

🚨 LANDMARK STUDY ALERT

The definitive 2019 network meta-analysis of antipsychotics just got a massive update—now with the muscarinic modulator KarXT (Cobenfy) and 78,000+ patients.

Clozapine's still best, Cobenfy looks strong, but side effects dictate choice 🧵

🤔 Background

For 70+ years, every approved antipsychotic blocked dopamine D2 receptors. We accepted the metabolic syndrome, weight gain, and EPS as the tragic cost of doing business in treating schizophrenia.

In 2019, the landmark Huhn et al. network meta-analysis (NMA) mapped out the efficacy of 32 oral D2 blockers, but we still lacked a novel mechanism of action.

Fast forward to 2024: the FDA approved xanomeline-trospium (KarXT, commercially known as Cobenfy), a first-in-class M1/M4 muscarinic agonist.

With MANY fast followers coming soon in the pipeline (I'm talking about nearly every company operating in the schizophrenia space with some kind of muscarinic modulator)...

...the field needed a new NMA to see how this entirely new mechanism stacks up against the old guard.

📋 Study Design

This is a mammoth NMA of 388 randomized controlled trials with 78,193 participants across 24 antipsychotics. Average trial duration was 6 weeks.

But here is the most striking methodological choice: they screened 5,117 Chinese trials.

Because of known quality issues in that literature, they directly contacted the authors.

Only 24 of those 5,117 trials could actively confirm proper randomization methods. They tossed the rest. That is an incredible level of rigor to protect the data's integrity.

For context - historically, previous NMAs of this size/caliber have tossed studies from China, despite some of these studies having incredible methodological rigor (see )

https://x.com/therealRYC/status/1984062469439156235